Definition
DMRPS measures the fraction of DMR data fields that pass a defined set of plausibility checks before submission, catching AI-generated or AI-assisted values that are statistically valid but physically implausible.
The Measurement Gap
Measurement Gap and Decision Supported
The Invisible Failure Mode: A wastewater utility implements an automated DMR preparation system that pulls data from LIMS and SCADA, computes statistical summaries, and populates DMR fields. The system reduces preparation time from 8 hours to 45 minutes per month. After 14 months of clean submissions, the state regulatory agency conducts a data quality audit and identifies that three months of ammonia data show impossible removal efficiencies: the AI system correctly computed monthly averages from LIMS data, but a sensor calibration drift in the influent sampler produced raw values that were too low, making effluent concentrations appear to exceed influent. The computed removal efficiency was negative, but the DMR field showed compliant effluent values because the system was programmed to report effluent concentration, not removal efficiency. A human reviewer would have noticed the implausible influent-effluent relationship; the AI system did not.
What Existing Metrics Miss:
DMR completeness checks verify that all required fields are populated. EPA's NetDMR system performs basic range checks. Neither catches the subtle plausibility failures that emerge when AI-assisted data processing introduces systematic errors that remain within normal numerical ranges but violate physical relationships between parameters. Without DMRPS, these plausibility gaps are invisible until a regulatory audit or enforcement action reveals them.
The Decision DMRPS Enables:
Is the AI-assisted DMR preparation process producing data that is not just numerically complete but physically plausible, and which specific parameter relationships require additional human review before submission?
Cross-References:
DMRPS connects to Regulatory Event Classification Accuracy (RECA, GMI-WT-029). Plausibility failures in DMR data may indicate events that should have been classified and reported as regulatory exceedances. Permit Condition Inventory Completeness (PCIC, GMI-WT-030) provides the regulatory framework context: DMRPS verifies data quality for the parameters PCIC tracks. DMRPS also connects to Prediction-Measurement Substitution Gap (PMSG, GMI-WT-021). If utilities are substituting AI predictions for measurements in DMR data, plausibility checking becomes even more critical.
Formula
Plain language: For each data field in the AI-assembled DMR, run three checks: (1) Does the value match the source data (LIMS/SCADA) in correct units and averaging period? (2) Is the value within the historical range for this parameter at this facility? (3) Is the value consistent with other parameters (e.g., pH and alkalinity should correlate; high BOD with low TSS is unusual)? The score is the fraction of fields passing all three checks.
Notation:
DMRPS = Fields_all_checks_passed / Total_DMR_fields
Where:
- Fields_all_checks_passed = DMR fields where: source match = TRUE AND historical plausibility = TRUE AND cross-parameter consistency = TRUE
- Total_DMR_fields = all data fields in the DMR for the reporting period
Units: Percentage per DMR submission. 100% = every field is plausible. Flag individual failures by check type for targeted review.
Proposed formula, pending validation pilots.